Mining visitors in Video Surveillance system

Mining visitors in Video Surveillance system Video Surveillance system records the events happens 24 hours, every day. Investigator needs to search video streams manually for evidence when an incident occurs. Log of interesting events is useful for investigator for analysis of video events. Face logging system consists of face image of individuals entering in surveillance area. The aim of system proposed in this paper, is to maintain log of visitor faces entering in surveillance area. System consists of face detection[4] and extraction, face log maintenance in database. For face detection Haar features are used and adaboost is used to get strong classifier for detection. Before applying Haar features, Probability based face mask prefiltering(PFMPF)[2] is used which can filter out more than 85% nonface images from an video frame. Use of PFMPF helps to reduce training period for face detection. Saving only the best face images of each target ensures that forensic analysis will not be overwhelmed with many redundant images of the same target. After face detection, quality filters are applied of face images to get best quality face image. Also log can be used to classify authenticated and unauthenticated faces entering in surveillance area if face recognition is implemented.